SoftwareSecrets

How to Use Claude Code: A Non-Coder's Guide (No Terminal)

Garrett Pierson

Claude Code is Anthropic’s AI coding agent, and you can run it from a desktop app that never asks you to open a terminal. You point it at a folder, describe what you want in plain English, and — by default — approve every change before it touches a file. It runs on a paid Claude plan, starting at $20/month for Pro.

If you’ve opened a Claude Code tutorial and closed it thirty seconds later because step one was pasting a command into something called PowerShell, that tutorial was written for developers. The desktop app is the same engine with buttons.

Before you start

Three things, and one of them costs money:

  • A paid Claude subscription. Pro is $20/month, or $17/month billed annually. Max starts at $100/month for heavy daily use. Anthropic’s docs are blunt about the free tier: “The free Claude.ai plan does not include Claude Code access.” (You can also authenticate with a pay-as-you-go Anthropic Console account, but a Pro subscription is the simpler path.)
  • A reasonably current computer. macOS 13 or later, Windows 10 (build 1809) or later, or Ubuntu 20.04+, with at least 4 GB of RAM.
  • A folder to work in. An empty one is fine — Claude Code can start a project from nothing.

On Windows you’ll also need Git for Windows installed for local sessions to run. Most Macs ship with it already.

Step 1 — Install the desktop app

Download the installer from Anthropic and run it like any other app. macOS and Windows both get a one-click installer; Linux is in beta via apt or a .deb package. Sign in with your Anthropic account, then click the Code tab at the top of the window.

The app has three tabs, and only one of them edits your files:

  • Chat — ordinary Claude conversation, no file access.
  • Cowork — a background agent that works inside its own sandboxed virtual machine while you do something else.
  • Code — the one this guide is about. It reads and edits files on your machine and shows you every change first.

The desktop app includes Claude Code, so there’s no Node.js to install and no command-line tool to set up alongside it.

If you do want the terminal version, the install is one line — curl -fsSL https://claude.ai/install.sh | bash on macOS, Linux, or WSL, and irm https://claude.ai/install.ps1 | iex in Windows PowerShell. Then run claude --version to confirm it worked. Claude Code also runs on the web, in VS Code, and in JetBrains IDEs. The desktop app and the terminal version share the same configuration, so a settings file you create in one applies in the other.

Step 2 — Pick your folder and your model

In the Code tab, choose Local to work on files on your own machine, then click Select folder and choose your project. There are other options — Cloud sessions that keep running after you close the app, SSH for a remote server — but Local is where you start.

Then pick a model from the dropdown next to the send button. You can switch it mid-project from the same place, so don’t agonize. If you want the reasoning behind model choice, we break it down in the best AI for coding.

Start with a small project you understand. Watching Claude Code work on something familiar teaches you more in ten minutes than an hour of tutorials.

Step 3 — Ask for one specific thing

Type what you want in the prompt box. The quality of what comes back tracks the specificity of what goes in, so give it something a colleague could act on without asking a follow-up question:

Add a contact form to the homepage with name, email, and message
fields. When someone submits it, show a thank-you message. Don't
change anything else on the page.

That closing boundary matters. Leave it off and an AI agent will happily “improve” four things you never asked about. We keep a full set of copy-paste Claude Code prompts organized by what you’re trying to do — planning, building, debugging, shipping.

You can also hand it context directly: type @filename to pull a specific file into the conversation, or drag images and PDFs into the prompt box. Screenshot a design you like and ask Claude Code to match it.

Step 4 — Read the diff before you accept

The Code tab starts in Manual mode, which means Claude proposes changes and waits. You get a diff view showing exactly what changes in each file, plus Accept and Reject buttons. Nothing on your computer changes until you click Accept.

Reject something and Claude asks how you’d rather proceed. You can also comment on specific lines in the diff, and it reads those comments and revises.

Three permission modes control how much freedom it gets:

Mode What happens Use it when
Manual Asks before every file edit or command Default — and where you should stay at first
Accept edits Auto-approves file edits You trust the task and want speed
Plan Proposes an approach, edits nothing Before anything big

In the terminal version, Shift+Tab cycles through these modes. Some accounts also get an auto mode that runs a background safety check and blocks risky actions.

Step 5 — Write down the rules once

Claude Code automatically reads a file called CLAUDE.md at the start of every session. That makes it the place to record decisions so you stop repeating yourself in every conversation. Ask for one directly:

Create a CLAUDE.md for this project. Record what this app is in one
paragraph, the rules you should always follow (never add new
libraries without asking, explain everything in plain English), and
how I test changes.

Keep it short. Every line in that file gets loaded into every session you ever run, so record the handful of decisions you’re tired of repeating and leave out the rest.

Step 6 — Use Plan mode for anything ambitious

Before a big change, switch to Plan mode and describe what you want. Claude works out an approach and shows it to you without editing a single file. You read the plan, correct the parts that are wrong, and only then let it build.

Plan mode is the habit worth building first. An hour of building on a bad plan costs far more than five minutes spent reading one. For larger projects, write the plan down first — our PRD template for AI app builders is built for exactly this.

What trips people up

  • Asking for five things in one message. You get five half-finished things and no idea which change broke what. One feature per request.
  • Accepting diffs without reading them. Skim every diff even when the code means nothing to you — that’s how you catch the moment a request to fix the login page also rewrites your database.
  • Never checkpointing. When the app reaches a state that works, ask Claude Code to commit it. A future bad change becomes a one-line rollback instead of a restart.
  • Expecting it to know what’s in your head. It reads your files, and that’s the whole of what it knows — your business, your customers, and the decision you made in the shower all have to be said out loud.

If you’re brand new to all of this, what vibe coding is covers the concept underneath the tool, and how to build an app with AI walks the full path from idea to shipped product.

The bottom line

Claude Code runs on a $20/month Pro plan, and the desktop app removes the terminal from the equation entirely. Install it, point it at a small folder, ask for one specific thing, and read the diff before you accept it — that loop is the whole skill. What it can’t do is pick a problem worth solving, price the product, or find your first customers. That gap is exactly what Software Secrets 2.0 was written to close, and the whole book is free.